Nishi Family › Compare › Deep Research and Answer Engines

Nishi vs the Field

Deep Research and Answer Engines — mechanically measured, liar-killed, sovereign.

Nishi vs Perplexity and DeepResearcher and Tongyi DeepResearch and deep-searcher

How this is scored. Every Nishi cell is measured: the generator reads the real organ source on disk and requires the implementing symbol to exist (no self-grading). Competitor cells record documented capability presence. Best=leads this axis, Yes=present, Part=partial, No=absent. This is capability presence, not depth or scale: the majors lead on index size and neural ranking. Nishi's genuine exceeds are the sovereignty / neutrality / determinism axes.
CapabilityNishiPerplexityDeepResearcherTongyideep-searcherNotes
Web fetch and bank pipelineYesYesYesYesYesRAG fetch loop; all fetch the live web
Lexical retrieval (BM25)YesYesYesYesYesSparse retrieval baseline
Dense vector retrievalYesYesYesYesBestdeep-searcher leads (Milvus / Zilliz vector DB)
Distributional embeddingsYesYesYesYesYesCo-occurrence semantic vectors
PPMI semantic matrixYesPartPartPartPartOurs explicit sparse; theirs neural
Trained embeddingsYesBestBestBestBestOurs PPMI-factorization toy; theirs neural leads
Multi-hop QA readerYesBestBestBestYesPretrained sovereign Qwen 1-shot + s0-window opt-in: on-protocol SQuAD 510 and Hotpot 505 (above mech-oracle 368) vs mech 250; MuSiQue 122 ties the composition wall (NAS-measured, multi-alias). A 2-shot surface-form variant won +104 on the single-gold standalone bench but did NOT transfer to the multi-alias on-protocol metric (SQuAD 427 vs 510) and was reverted (confirmed by nx_reader_ruler_adversary_gate 9/9). Evidence-grounded: SQuAD human F1 86.8 / LR-baseline 51.0 per fetched primary source arxiv:1606.05250 (nx_evidence_verify_gate 7/7); published neural SOTA (approx 900 permille) still leads
QA F1 / EM scoringYesYesYesYesYesSQuAD-style measured evaluation
Findings ledger (compounding memory)YesYesYesYesPartPersistent provenance ledger
Corpus semantic retrievalYesYesYesYesBestdeep-searcher leads
RL-trained research agentNoYesBestBestNoDeepResearcher / Tongyi GRPO-trained lead; Nishi none
LLM synthesis / reasoningNoBestYesYesYesPerplexity leads (polished synthesis); Nishi extractive only
Adversarial refutationNoYesBestYesNoDeepResearcher cross-validates across sources (leads); Nishi has a GATE-PROVEN prototype -- evidence-grounded claim verify (confirm/discredit vs a fetched primary source, nx_evidence_verify_gate 7/7) + metric-ruler adversary (nx_reader_ruler_adversary_gate 9/9) -- NOT yet wired into the live researcher (census correctly holds this absent until wired)
Neural reranker (cross-encoder)NoYesYesYesPartTrained rerank; Nishi lexical + sparse only
Multimodal report outputNoBestNoNoNoPerplexity leads (charts / images); Nishi text only
Served research API / productNoBestPartPartYesPerplexity leads (product API); Nishi organ-only
Sovereign -- own substrate, TLS, index; zero OpenAI / cloud dep; 0 dollar per queryBestNoNoPartNoAll use hosted LLMs; Tongyi open-weights partial; Nishi fully sovereign
Integer-deterministic, crash-resumable, no link-rotBestNoNoNoNoReproducible; peers are non-deterministic LLM pipelines
coverage 666/1000EXCEEDS 2PRESENT 10ABSENT 6
Honest verdict. Nishi has broad, sovereign search-engine presence — a real organ behind every core ranking and crawl axis, built bits-up in one language. It is not yet a scale or neural-ranking rival: the majors lead on index size (billions to trillions vs a Common-Crawl-seeded slice) and trained rerankers. Nishi's genuine exceeds are structural — owning the whole stack, neutrality by construction, and a zero-JS integer-deterministic result page. The climb is a bigger index and a trained reranker.

Generated by nx_swcompare_matrix (sovereign NishiLang organ) from knowledge/compare/deepresearch.matrix — every Nishi cell verified against real organ source on disk. Zero JS, zero trackers.